This is an example application that lets you respond to a Twilio Voice call using Twilio ConversationRelay and Langflow.
You will need:
- a Twilio account with a voice-enabled phone number
- to follow the steps to enable ConversationRelay in your Twilio Console
- a Langflow installation
- a way to tunnel requests to your local machine, like ngrok
First clone this repository from GitHub:
git clone https://github.com/langflow-ai/langflow-twilio-voice.git
cd langflow-twilio-voice
Install the dependencies:
npm install
Copy the .env.example file to .env.
cp .env.example .env
We'll fill in the entries in the .env file as we continue.
You will need a Langflow flow to handle the messages from the voice call.
It will be interacting with a live voice call, so it should aim to respond quickly. It is also a live conversation, so you will want to ensure that there is built in memory.
You can try using any of the template flows in Langflow, or for a basic chat bot with memory try importing and building upon ./langflow/hackathon_bot.json.
Once you have created the flow you will need to get the base URL for your Langflow flow. The default for Langflow installed by pip/uv is http://localhost:7860 and for Langflow desktop it is http://localhost:7868. Enter this as the LANGFLOW_URL in .env.
You will also need the ID of your flow. You can find this in the examples under Publish -> API access. The URL looks like http://localhost:7860/api/v1/run/${YOUR_FLOW_ID}. Enter this in the .env file as LANGFLOW_FLOW_ID.
If you have enabled authorization on your Langflow installation, you will need to generate an API key. You can do so in the settings. Add the API key in .env as LANGFLOW_API_KEY.
Your Langflow bot will be communicating by voice over the phone, so your system prompt should encourage it to be concise and clear. Consider using something like this in the system prompt:
This conversation is being translated to voice, so answer carefully and concisely. When you respond, please spell out all numbers, for example twenty not 20. Do not include emojis in your responses. Do not include bullet points, asterisks, or special symbols.
(Inspired by this example repo from Twilio's Kelley Robinson)
If your flow is capable of streaming responses, you can take advantage of the lower latency by setting LANGFLOW_STREAMING="true" in the .env file.
To connect Twilio to your locally running application, you will need a tunnel like ngrok that supports web socket connections.
This application runs on localhost:3000 by default, or whichever PORT number you set in .env.
Set up your tunnel to point to the locally running application and enter the tunnel domain in .env as TUNNEL_DOMAIN.
You will need to configure your voice capable phone number to connect to the application. Under the voice configuration settings of the phone number, configure it with a webhook. When a call comes in the webhook URL should be a POST request to https://${TUNNEL_DOMAIN}/voice.
You can run the application with:
npm start
Now dial your Twilio phone number and you can chat with your Langflow flow.
You can also run:
npm run dev
This runs the application in watch mode, so it restarts if you make any changes. Be aware this may cause issues with live calls losing connection.